Do you need multiple Application Servers?

[Jack_Crouch]

We are on XE now, but I am studying up on the idea of setting up a 9.0 environment. It would be 8.98.3.1, Deployment VM Windows2008x64SP2 SSE, Enterprise/Data System i V5R4. WAS ND 70 on System i.

When talking to people about using OAS or WAS on intel, it seems "assumed" I would have at least 2 intel servers "for redundancy". I am not sure if they are saying that for hardware redundancy or software redundancy reasons.

Since I am thinking of putting WAS on one System i, I have this question:

I will install the ND version of WAS (because I can - that is what is in Oracle eDelivery). I wonder if I need to create multiple application servers (vertical cluster I guess) underneath that install or not? We will have 600 concurrent users. I am not clustering for hardware redundancy (since it is all on one System i). But I wonder if in the WAS world, if maybe it is not uncommon for an Application Server instance to “hang up” and require recycling. If that is true then I guess you might want multiple Application Servers just to avoid bringing the entire company down during a recycle?

Or maybe you want multiple Application Servers because it keeps the number of Java threads down and the "garbage collector" has an easier time?

 

[klwong]

For 600 concurrent users, I would have several JAS servers to handle the load. By "several," I mean >=5.

I don't know anybody who recommends handling more than 200 users on a single instance of WebSphere or the Oracle alternatives.

 

[tiradoj]

Hello Mr. Crouch.

Long time no chat. Joe Tirado here. I was following up on my own post a couple days ago and thought I would make a not so quick comment which may only be marginally related to your question and maybe it is worth two cents maybe not.

Short answer, I would say yes, for the number of concurrent users you have and to avoid complete PROD down time.

I haven't been asked to do a WAS configuration in a while (and perhaps have not been the best promoter of WAS recently, but only because I have been using OAS so much) so I could not offer a "hard analysis and benchmark" comparison between the two app servers, but I see "redundancy" from a practical and troubleshooting perspective as always a good thing if a customer can afford-arrange it.

With only 45 concurrent users (300 defined) in PRODUCTION on our 8.12 and 89820 instance at one of my customers, admittedly a "small" user instance (important though as they do run Payroll in JDE) and the box CPU is never tapped out but PROD for all is on one "server".

Customer is considering implementing Employee and Manager self service would could significantly affect the number of users on the app servers, if only for the relatively short enrollment period, but could be up to 1500 or more users hitting the box-server-virtual at the same time so our scenario could change quickly.

We have a separate TEST web server we could failover to (not a true failover, more of a "hey use this URL instead) and use for accessing PROD now if we had to--but has 1/4th or less the power of the PROD server and supports all the other non PROD environments--theoretically we could rely and I believe are also relying on a VMware image of PROD but have not tested the fail and recovery scenario in a test fail--despite my best effort multiple recommendation-warnings to customer to do so. Too many other priorities I guess but I am waiting for the other shoe to drop.

As an aside, in case you are thinking of piling other JDE related apps on your WAS, we are also implementing the eRecruit portal application for JDE (Human Resource) for requistions from the Internet which called for it's own application server of course (per Oracle MTR) with Oracle Portal Server and because of OS restrictions on the release we are working with we we're limited to a 32 bit machine was advised we could be expected to comfortably support "roughly" (from inquiries with other customers and several repeated requests from Oracle direct) 200 light to medium web users of eRecruit.

How anyone could say this was a good measure with no documentation...

but I digress.

Ultimately we have no idea how many Internet users will access the eRecruit application (although I was told the company receives up to a few hundred resumes a week, I think that is a bit overestimated, but who knows.)

No one might access or the server could be deluged but my customers are big fans of virtualization and I am a pretty good fan too, where workable and tested.

Again this many be marginally related to your question but I suspect that WAS must in some important respects should have some of the same challenges with regard to scalability and failover, but again I could be off not knowing the very latest specifics on WAS release's most current release.

This reply was wayyyy too long.

Anyway
Glad to see a fellow JDE gladiator on the list.
Joe Tirado

 

[Jack_Crouch]

Greetings Joe... and thanks guys... So I think the 200 (ideally 100) users per application server is a software "application server" type of limitation. So I should expect to eventually create 6 application servers on my i. And I need to figure out how to use WAS ND to distribute the load across those. I assume I would have one HTTP server instance involved with those 6.

 

[shearerj]

That is correct. One HTTP server would provide the front end to load balance between the WAS servers.

 

[brother_of_karamazov]

[ QUOTE ]
That is correct. One HTTP server would provide the front end to load balance between the WAS servers.

[/ QUOTE ]

Or more for redundancy.

 

[gregglarkin]

[ QUOTE ]


Or more for redundancy.

[/ QUOTE ]

Redundancy is good. Redundancy is good.

For the environment we are building, we will be using F5 Big IP switches to load balance the web and application servers.

- Gregg

 

[brother_of_karamazov]

[ QUOTE ]
We are on XE now, but I am studying up on the idea of setting up a 9.0 environment. It would be 8.98.3.1, Deployment VM Windows2008x64SP2 SSE, Enterprise/Data System i V5R4. WAS ND 70 on System i.

When talking to people about using OAS or WAS on intel, it seems "assumed" I would have at least 2 intel servers "for redundancy". I am not sure if they are saying that for hardware redundancy or software redundancy reasons.

Since I am thinking of putting WAS on one System i, I have this question:

I will install the ND version of WAS (because I can - that is what is in Oracle eDelivery). I wonder if I need to create multiple application servers (vertical cluster I guess) underneath that install or not? We will have 600 concurrent users. I am not clustering for hardware redundancy (since it is all on one System i). But I wonder if in the WAS world, if maybe it is not uncommon for an Application Server instance to “hang up” and require recycling. If that is true then I guess you might want multiple Application Servers just to avoid bringing the entire company down during a recycle?

Or maybe you want multiple Application Servers because it keeps the number of Java threads down and the "garbage collector" has an easier time?

[/ QUOTE ]


More on load balancing:

http://jeffstevenson.karamazovgroup.com/2008/09/load-balancing-at-all-levels-in.html

 

[Alex_Pastuhov]

Our SSO software offers round-robin load balancing as a free feature (i.e.: you do not need to have it licensed to use this feature). To get the actual software, one just needs to e-mail me...